mcp-architecture-skill

Reduce MCP server token usage by implementing progressive disclosure so agents discover tools on-demand, and use resource-based data access via URIs to keep large payloads out of the context window.

Progressive disclosure in Model Context Protocol architecture is a pattern where agents discover tools on-demand rather than loading all tool definitions upfront, drastically reducing initial context size and token overhead.

Refactor an existing MCP server for token efficiency by migrating to progressive disclosure for tool discovery and routing large data payloads through external resources accessed via URIs, minimizing upfront context loading.

Manage large data payloads in MCP by using resource-based data access, storing data externally and providing agents with URIs to retrieve specific information only when needed, keeping context clean.

This MCP architecture pattern supports code execution by designing atomic and composable tools, allowing agents to write scripts that chain multiple operations together for complex data processing workflows.

Loading all MCP tools upfront creates severe token inefficiency and scalability issues, often consuming tens of thousands of tokens, which necessitates adopting on-demand discovery and external resource management patterns.